Research map / Detection and segmentation / CNN features
DetNAS: Backbone Search for Object Detection
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Search the backbone on object detection rather than ImageNet classification | DetNAS versus ClsNASNet · COCO FPN, COCO RetinaNet, VOC FPN, and VOC RetinaNet | detection mAP: 35.1 -> 36.6; 31.2 -> 33.3; 78.5 -> 81.5; 76.5 -> 80.1 | Helped |
| Add ImageNet pre-training to the detection-search pipeline (ablation) | DetNAS · COCO FPN, COCO RetinaNet, VOC FPN, and VOC RetinaNet | detection mAP: 35.9 -> 36.6; 32.8 -> 33.3; 81.1 -> 81.5; 79.9 -> 80.1 | Helped |
| Search the backbone instead of using the hand-crafted ShuffleNetv2-20 | DetNAS · COCO FPN, COCO RetinaNet, VOC FPN, and VOC RetinaNet | detection mAP: 34.8 -> 36.6; 32.1 -> 33.3; 80.6 -> 81.5; 79.4 -> 80.1 | Helped |
| Use the detection-searched architecture rather than random architectures | DetNAS · COCO FPN, COCO RetinaNet, VOC FPN, and VOC RetinaNet | detection mAP: 35.6 ± 0.6 -> 36.6; 32.5 ± 0.4 -> 33.3; 80.9 ± 0.2 -> 81.5; 79.0 ± 0.7 -> 80.1 | Helped |
| Scale DetNASNet from 1.3G to 3.8G FLOPs (ablation) | DetNASNet · COCO minival, FPN, 1x | mAP: 40.2 -> 42.0 | Helped |
| Replace the DetNASNet backbone with ResNet-50 | DetNASNet (3.8) · COCO minival, FPN, 1x | mAP: 37.3 -> 42.0 | Hurt |
| Replace the DetNASNet backbone with ResNet-101 | DetNASNet (3.8) · COCO minival, FPN, 1x | mAP: 40.0 -> 42.0 | Hurt |
Closest papers by meaning
- Joint-DetNAS: Upgrade Your Detector With NAS, Pruning and Dynamic Distillation (CVPR 2021 · similarity 0.92)
- SM-NAS: Structural-to-Modular Neural Architecture Search for Object Detection (AAAI 2020 · similarity 0.91)
- DetKDS: Knowledge Distillation Search for Object Detectors (ICML 2024 · similarity 0.88)
- MI-DETR: An Object Detection Model with Multi-time Inquiries Mechanism (CVPR 2025 · similarity 0.86)
- Order Matters: Competition-Guided Query Ordering for RNN-Based Object Detection (NeurIPS 2026 · similarity 0.86)
- MS-DETR: Efficient DETR Training with Mixed Supervision (CVPR 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.